Rights-cleared data for AI
Tributary sources rights-cleared, real-world data from many independent origins and delivers it to AI labs and data vendors as one clean, model-ready stream.
01
Creators, communities, enterprises and field operators — each holding a sliver of the real world that models have never seen.
02
Consent, licensing and provenance are established before a single byte moves. Nothing scraped, nothing ambiguous.
03
Normalized, de-duplicated, documented and shipped in the format your pipeline expects — ready for training or evaluation.
The middle of the supply chain
Supply
Tributary
Clear · curate · contract · deliver
Demand
For sources
For buyers
How it works
A single, accountable pipeline between the world's data and your models. You get one counterparty, one contract and one consistent standard — no matter how many sources sit upstream.
We find and recruit the people and organizations who hold the data you need — or commission it from scratch when it doesn't exist yet.
Every record is rights-cleared: explicit consent, a clean license chain and a provenance record that travels with the data.
We normalize, de-duplicate, filter, redact PII and annotate to your spec, with QA sampling at each stage.
Datasets ship with datasheets, manifests and the splits you asked for — via secure transfer to your bucket, on your schedule.
What flows through
Off-the-shelf where it exists, commissioned where it doesn't. Every dataset ships under the same rights, quality and documentation bar.
Expert-written long-form, multilingual corpora, domain documents and instruction data.
JSONL · Parquet
Conversational and read speech across languages, accents and acoustic environments.
WAV · FLAC + transcripts
Licensed photography and documents with captions, bounding boxes and segmentation.
PNG · JPEG + COCO
Egocentric, instructional and in-the-wild footage with temporal annotations.
MP4 · WebDataset
Robotics teleoperation, multi-sensor rigs and telemetry for physical-world models.
MCAP · HDF5 · RLDS
Held-out, contamination-controlled sets and expert rubrics for benchmarking and red-teaming.
Private · never-trained-on
Trust & provenance
The value of training data is only as durable as the rights behind it. We build every dataset so you can answer the hard questions — where did this come from, who agreed, and on what terms — record by record.
{
"record_id": "trb_9f3c…a21e",
"modality": "speech",
"source": {
"type": "contributor",
"consent": "explicit, training + eval",
"consent_ts": "2026-08-14T16:02:11Z"
},
"license": {
"grant": "non-exclusive",
"scope": ["training", "evaluation"],
"chain_verified": true
},
"processing": [
"pii_redaction@v3",
"dedup:minhash",
"transcript_qa:sampled"
],
"withdrawable": true
}Your model is only as good as the sources that feed it.
Get in touch
Whether you're training a frontier model or sitting on data the world's models need, tell us what you're working on. Every inquiry is read by our team.
hello@tributarydata.ai